Anthropic’s pricing page lists a Free tier at $0 and describes it as “Free for everyone”. It is not a seven-day trial, and not a demo stripped of the useful parts.
The harder question is where the free tier stops. The answers are split across two official pages, the pricing comparison table and a help centre note about a five-hour reset, so this guide puts them side by side.
What the free tier is
The pricing page splits personal plans into Free, Pro and Max, with Team and Enterprise above them. Free is $0. Pro is $17 a month billed annually, as $200 up front, or $20 billed monthly. Max starts at $100.
What separates Free from a trial is that no expiry date is published, and features are not withheld to force an upgrade. The boundaries are usage and models, not time.
Included features, and the gaps
The Yes rows on the Free column are broader than most people expect: chat on web, desktop and mobile; web search; creating files and running code; memory across conversations; connectors to outside apps; Artifacts; voice mode; incognito chats; Skills; and desktop extensions.
The No column decides your workflow. Absent from Free: Claude Code, Projects, Research, @Claude, Claude in Chrome, the Microsoft 365 and Outlook integrations, Design/Slides/Docs, enterprise search.
What Sonnet and Haiku only means
The model row is the hardest limit. The published table lists Sonnet and Haiku as available on Free, and Opus and Fable as not. The context window on Free is 200k, the same as Pro and Max.
So the full 200k context comes without the two strongest models. Writing, reading long documents and small scripts sit inside what Sonnet and Haiku cover.
Getting in and saying something
Three doors are published: claude.ai in a browser, the desktop app for Mac or Windows, and the iOS or Android app. One account covers all three.
Two conditions come with them: a supported location, and being at least 18. The full list is not on that page.
The help centre’s advice on how to open is plain: talk to Claude the way you would talk to a coworker or a friend, begin with simple questions, be specific, iterate, and try different kinds of tasks.
Example: three prompts to copy
None of these are vendor samples. They exist to make a limited quota count, and are meant to be pasted as they are.
Case one: hand over a spreadsheet instead of counting rows yourself. Asking for an analysis is too vague.
This is my [bills / sales records / attendance sheet].
Answer three questions:
1. Which category has the largest share, and what are the top five?
2. Are there outliers, values that jump up or down suddenly?
3. If I want to save 500 a month, which items are most realistic to cut?The third question is the one worth copying: the first two are statistics you could read yourself.
Case two: a fixed three-question pass over a long document.
Tell me in five sentences what this document says, and who gains or loses.
Which clauses or conclusions would hurt me later if I ignore them now?
If I only have ten minutes, which pages should I read?The second question is the core of it: the model will read a hundred pages without getting tired.
Case three: make Claude the person who picks holes in your plan, turning a vague worry into specific questions.
I am planning to [quit for freelance work / enrol my child in this course / buy this car].
My reasons are [...].
Play a rational but sceptical friend and list five traps this decision could fall into,
with a signal for each trap that would show it is actually happening.All three spend ordinary chat quota, and none need Claude Code or Projects.
How the free quota actually works
The pricing FAQ is explicit. Every plan has usage limits that reset on a rolling five-hour session window, paid plans add weekly limits, and your activity on web, desktop, mobile and Claude Code all draws from one pool. As for how many messages that is, the published position is that there is no fixed count.
What is published is the relative scale: Free covers everyday questions, Pro gives at least five times the usage per five-hour session, Max gives five or twenty times Pro. The help centre states the free limit as a session-based limit resetting every five hours, with the number of messages varying with demand.
Usage credits read No on Free, which is how paid plans keep working after a limit.
When paying starts to make sense
Any one of three signals is enough to run the numbers.
First, you start caring about that five times. The limit is a window rather than a monthly allowance, so it returns within five hours; the test is whether the interruption costs more than the subscription.
Second, you need a feature the table marks No. Claude Code is No on Free while the FAQ says it is included in all paid plans, and Projects and Research are paid-only.
Third, you need a stronger model. Opus and Fable are both No on Free. Pro reaches Opus; Fable draws on usage credits.
If your usage is occasional questions, document reading and writing, the free tier will not be the constraint. The breakdown by tier is in the plan comparison.
Frequently asked questions
Is Claude free? Yes. The pricing page lists Free at $0 and describes it as “Free for everyone”, with no expiry date published.
Which models can I use for free? The published table lists Sonnet and Haiku as available and Opus and Fable as not, with a 200k context window.
How many messages do I get every five hours? Not published. The pricing page states there is no fixed message count, and the help centre says the number varies with demand.
Does the free tier include Claude Code? No. The comparison table lists Claude Code as No on Free, while the pricing FAQ says it is included in all paid plans.
Do I need a card to start? Nothing in the official material describes a payment requirement for the free tier.
Where should I read next? The product path is in the Claude getting-started guide; the free-versus-paid boundary has its own breakdown; and to call the API directly, start with the API getting-started guide.
Start on Free, watch for the limit rather than guessing at it, and decide about paying when you hit it.